Let the bots battle to their elimination. They won’t help you either!

robot jim/Adobe Stock

Editor’s note: This Spend Matters PRO post has been unlocked to continue our exploration of how to get the most value out of your procurement technology and to provide a sample of the depth we go into in our PRO material.

In our piece titled “Have You Mastered the Procurement Basics yet? we explained how AI won't save you if you haven't covered the basics, because even the best AI needs Big Data (which is likely bigger than you have) to learn from and considerable training to be effective.

Furthermore, while we briefly mentioned Automation, we didn't discuss it in depth, nor did we discuss it outside of AI. Nor did we discuss Robotic Process Automation, or Bots, which various vendors are vigorously vending to the virtuous, who will not gain from their implementation and, in fact, may lose dearly.

This is because, despite all the marketing claims to the contrary, at least today, Bots are nothing more than packaged automation scripts with simple rules and context that can be used to automate basic tasks, tasks that sometimes are less complex than those that can be figured to be handled automatically in more mature platforms.

It's also because Bots are NOT INTELLIGENT. Vendors want you to associate Bots with AI and believe that the actions of the bot will be at least as useful and intelligent as actions that your intelligent and trained personnel take, but that is not likely. A bot is just going to automate a process that is now driven manually, with no more intelligence than the human, and in fact, even less if the rule definition does not permit a rule as expressive or complete as a rule a human would follow.

For example, let's say right now Procurement is still "processing" all invoices for payment, and the rule is all invoices that m-way match to 99% or more accuracy that is under 50,000, from a known supplier in good standing, doesn't have any suspicious flags, and looks OK is "OK to pay.” If the best the software permits is:

  • verify under X dollars
  • verify two-way match
  • verify supplier not blocked
  • verify payment information has not changed
  • verify items have been ordered before

That is good and might seem good enough for the average organization, but is it?

Let's take the capabilities one-by-one:

  • A dollar limit is how most companies approach approvals, but it's actually among the riskiest methods — as soon as someone knows the limit, they will split invoices to slide just under it if they don't want scrutiny.
  • A two-way match is great — PO, invoice — but the match should be between whatever documents are available; contract addendums, statements of work, returns, etc., not just PO and invoice.
  • Just because a supplier is not yet blocked, that doesn't mean they are OK to pay — they might be extremely high risk, the goods may not yet have been verified as OK, etc.
  • It's good to verify payment information against the last time, but was it verified properly the first time?
  • This is good, as it makes sure the items are at least typical of what the organization buys, but doesn't mean this order, or invoice, was valid (especially if the items represented what was a one-time buy for an event or a stopgap during a contract switchover between two suppliers).

Moreover, and this is the key:

  • Such a system won't catch if the quantities are reasonable — if the organization typically orders 10 units of something, but the invoice has 100 (and just fits under the auto-approval threshold), this could be an example of fraud (external, internal or collusion).
  • Such a system won't catch if a volume discount or credit for a defective shipment has been applied — and if not caught before a contract is terminated, good luck ever capturing it.
  • Such a system won't catch if this is the second order to a supplier in two years, followed by a first order a week ago for only 4,000 (whereas this order was 40,000), which was a fraudulent invoice for goods not coming that is from an entity that is pretending to be the former supplier and slipped the first invoice through manual verification as everything but the actual bank account number was the same as the real supplier's data (including the actual bank address).
  • And other fraudulent or risky scenarios (where the risk module indicates a 95% of a supplier going bankrupt and/or missing the next shipment that they have requested pre-payment for), etc.

This is just one of many examples where Bots will be sold to reduce tactical burden and increase efficiency, but actually worsen results (as this bot would lead to more risky and fraudulent payments) when the best bot would be one that did more pre-approval analysis and queued invoices into a "perceived OK to pay" and a "perceived not OK to pay" queue where an expert could do a quick scan and batch approve — capabilities better platforms will already contain.

Another example where Bots are being sold to reduce tactical burden is to simplify data integration between platforms. This is totally fine, but if the platforms have APIs, SFTP and polling and configuration options — shouldn't the integration be possible without spending more bucks on a bot?

An emerging use is in the area of "cognitive" sourcing where you just hand your categories over to the platform for automatic sourcing. This is great in theory, and might work just great 90% of the time, but that last 10% of the time can get you just like that 2% of fraudulent invoices that will slip through on top of the 2% to 20% of invoices that might not be okay-to-autopay without a final human check.

For the right non-strategic, non-high value categories where the suppliers are known, the products are known or commodity, the market rates are known, the demands are known, the delivery times are reliable, and the only terms and conditions are the standard Ts and Cs the supplier agreed to in order become an approved supplier. It is true that the platform can automate just about everything including an award recommendation, but an award should still be reviewed for the same reason a considerable percentage of "perceived OK to pay" invoices should still be quickly eyeballed by a human expert — the system does not know everything (it only knows the data it is given), cannot model everything, and until it is trained (and learns an appropriate rule), it cannot know how to properly respond to a new situation that might affect an award (trade injunction with the supplier's country, tax hike, internal damage claim, etc.).

But, again, the real question is, why do you need a (third party) bot? A good platform should know the approved suppliers and products, track your demands, integrate with market price feeds (and your prior quote database) and know the expected market price ranges (compared to what you are paying now), the organizational demands, and the frequency of those demands. In any decent, modern, platform, it should be possible during the definition of an event (which a modern analytics platform should also recommend based upon demand, the price the organization is paying not under contract, the current market price, and expected demand), the buyer should be able to say auto-invite improved suppliers, auto-initiate the auction for x days, auto-compute the recommended award, and notify me in x days with the recommendation for one-click approval.

Why do you need a bot? (You don't, but vendors obviously want to charge you for something that's nothing more than a small collection of rule-based packaged scripts that should be included for free, as we've already explained multiple times, they are not intelligent, and you are not getting anything close to the AI they hope you think you are getting when they inflate the bill.)

In other words, let the Bots battle it out until they eliminate themselves. The necessary automation capability (whether it's called configuration, automation, RPA, bot, etc.) should be included in the platform to the extent it makes sense to do so, and you shouldn't need more dumb scripts to get good results — it's all marketing, and until it's otherwise, just don't care.

What you need, as we explained last time, is a good command of the basics and proper support for the basics. And a platform that helps you identify what you need to spend your time on, and what you can configure rules to automate once you have figured out what the right rules are to ensure value and prevent loss.

Share on Procurious

Discuss this:

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.